Nanubala Gnana Sai

Product Engineer

Bengaluru, Karnataka, India3 yrs 11 mos experience

Most Likely To SwitchAI Enabled

Key Highlights

Nearly four years of experience in machine learning.
Contributed to significant open-source projects.
Mentored peers in advanced AI techniques.

Stackforce AI infers this person is a skilled AI and Machine Learning engineer with a focus on logistics and optimization.

Contact

Skills

Core Skills

Ai SafetyKnowledge GraphsAws BatchPandasNode.jsReinforcement LearningMachine LearningMulti-objective OptimizationEdge Ai

Other Skills

CC++Computer VisionContent WritingData AnalysisData ScienceDeep LearningElastic SearchEvolutionary AlgorithmsFlaskGPUGitInternet of Things (IoT)Large Language Models (LLM)Low Level Programming

About

I believe a majority of world problems can be solved by AI. The idea of a modern era powered by it is fascinating to me. I find myself imagining a world where issues such as terrorism, water shortage, climate crisis are solved owing to this. I currently serve as a research-oriented software engineer at Shipsy. My work tackles diverse industry challenges in logistics such as hyperlocal deliveries, territory optimisation, geocoding, courier-express-parcel (CEP) deliveries, and roster management. I'm looking to broaden my horizon and connect with industry-leaders, think-tanks & researchers to advance my vision for AI's potential and further my growth in this dynamic field. I bring nearly four years of experience in machine learning, highlighted by meaningful contributions to open-source projects, productive collaborations with international peers, and published research paper(s) (happy to add more to this :)). Core values: Visionary, Ownership, Reliance, Discipline. ▪️Python, C++, JavaScript, Node.js ▪️TensorFlow, PyTorch, scikit-learn, pandas, geopandas ▪️AWS, PostgreSQL, MongoDB, Redis, Docker, Jenkins ▪️Elastic Search, Newrelic

Experience

3 yrs 11 mos

Total Experience

3 yrs 11 mos

Average Tenure

3 yrs 11 mos

Current Experience

Eleutherai

Research Fellow

Aug 2025 – Present · 10 mos · Remote · Remote

This project tackles "AI safety navigation problem" by building a knowledge extraction system to prevent critical interventions from being lost in the flood of new research. Mentored by Martin Leitgab.
As a team, we:
Extracted intervention chains from 600+ papers in the Alignment Research Dataset (ARD).
Built a knowledge graph with 8,000+ nodes and 6,000+ edges.
Validated extraction quality by comparing LLM-generated graphs against human-annotated ground truth from a curated holdout set.
Developed a 6,000+ line open-source codebase with contributions from 10+ collaborators.
My direct contributions:
Designed the end-to-end pipeline, making key architecture calls on embedding selection (BAAI/bge-en-large-v1.5), graph database choice (FalkorDB), and pipeline structure.
Resolved bottleneck in LLM querying by implementing a Batch API framework, expanding rate limit and cutting cost by half.
Contributed to the creation of a manual graph validation dataset.
On-going Work & Future Plans:
Extending the system with semantic compression using LSM-Tree methods to merge duplicate concept nodes at scale
The aim is for a workshop paper and LessWrong publication.

AI SafetyKnowledge GraphsScalable OversightLarge Language Models (LLM)

Shipsy

4 roles

Senior Software Engineer

Promoted

Apr 2025 – Present · 1 yr 2 mos

Software Engineer L2

May 2023 – May 2025 · 2 yrs

Mentored intern to eliminate solver idle time of routing service achieving 98% runtime reduction, potentially saving ~ $1.6K annually
Implemented an intelligent caching system to capture hub metrics, reducing response time per API call while ensuring data freshness through strategically timed cron jobs.
Implemented an intelligent caching system to capture hub metrics, reducing response time per API call while ensuring data freshness through strategically timed cron jobs.

New RelicProgramming Languages

Software Engineer

Jul 2022 – May 2023 · 10 mos

Territory Optimisation:
Inefficient existing processes of creating sales territories led to uneven workload distribution and increased reliance on freelance workers to meet demand. Engineered a K-Means based clustering algorithm to produce sale areas; used AWS Batch to offload compute and to scale with load.
Improved system reliability from 82% to 99.98% using retry mechanism on AWS Batch
Routing:
Scaled routing algorithm to handle 1000-point problems (10x increase) while maintaining a 15-second P95 response time, supporting 246k daily calls.
Reduced service error rate (5xx response) to 0.002% (20 per million) by developing a comprehensive test-suite in pytest in collaboration.

AWS BatchPandas (Software)pytestParallel ComputingNode.jsProgramming Languages+1

Software Engineer Intern

Jan 2022 – Jul 2022 · 6 mos

Extended vehicle routing algorithms to meet client-specific constraints such as: prioritizing decaying orders, clubbing orders of nearby locations, multi-day trips, etc.
Enhanced system monitoring by connecting AWS Kinesis Firehose to log data on OpenSearch.
Improved model explainability by reporting drop reasons of consignments during allocation.

OpenSearchElastic SearchFlaskOperations Research

Google summer of code

GSoC Mentor

May 2022 – Sep 2022 · 4 mos · Remote · Remote

Mentored implementation of critical reinforcement learning algorithms to expand mlpack's reinforcement learning modules.
Guided development of:
Proximal Policy Optimization (PPO)
Twin Delayed DDPG (TD3)
Hindsight Experience Replay

Reinforcement LearningC++Machine LearningDeep Learning

Google summer of code

GSoC'21 Student@mlpack

May 2021 – Aug 2021 · 3 mos

This project aims to add optimizers, expand the test framework and make the library more accepting of multiobjective problem suites.
Implemented MOEA/D-DE, a novel multi-objective evolutionary algorithm. Introduced the ZDT test suite to the library and measured the convergence of the said algorithm.
Systematically benchmarked speed and convergence against the traditional NSGA-II algorithm and reported 32x times faster performance with 20% higher relative convergence.
Demonstrated practical applications of the algorithm via interactive Jupyter notebooks in portfolio balancing and rocket injector design problem. Plotted 3D interactive graph to visualize the solution set. Used Unity WebGL to showcase a demo rocket injector surface and optimal design variables.

PythonC++Machine LearningMulti-objective OptimizationEvolutionary AlgorithmsData Science

Gmac intelligence

AI/ML Intern

Aug 2020 – Nov 2020 · 3 mos · Andhra Pradesh, India

The core idea was to build a framework that would allow users to perform inference on edge. The framework equips the user to chain multiple models together and is agnostic of the machine learning task.
Key highlights:
Built Native C++ library to perform high-speed inference on ARM mobile devices.
Successfully tested a sample image classification code on Xiaomi Redmi7.
Harnessed GPU using the OpenCL library, reported 3x boost in inference speed.
Utilized DSP, XNNPACK and NNAPI delegate to benchmark the performance boost.

Internet of Things (IoT)C++Machine LearningGPUEdge AITensorFlow+1